TL;DR: There may finally be a new server architecture! Should you care?
Last week, I found myself split screen streaming VMworld and the LF Open Network and Edge Summit (ONES) – check out my TFIR Recap! While they’re very different shows: a proprietary vendor and an open source community, they’re also remarkably the same. And that says a lot about the scattered state of our industry.
Both conferences were marked by a caravan of large vendors eagerly lined up to support the mission of each conference. In fact, it’s the same vendors supporting BOTH the proprietary and the open source software efforts. Plus we need to throw in Nvidia (acquired Mellanox) who seems to be all over this space too. And while the voice of the hyperscale providers was missing in these conferences, their presence is clearly being felt. Wow, that’s a lot of interest.
So why would major vendors dog pile into parallel proprietary and open events? It means while the use cases that we’re trying to solve as an industry are well understood BUT the technical solution is not.
What we do have happening is the major infrastructure platforms are so satisfied with what they’re doing and their plans and their mission to create walled gardens, that they feel very little need to participate collaboratively inside of these more general industry venues. And don’t make any mistake, from that respective ecosystems, VMware is as open a platform as the Open Networking and Edge Summit presents with the Linux Foundation. They have different objectives, but they both are vibrant ecosystems with active communities providing value on top of the components of their core platforms.
So the thing that’s interesting to me is not as much the differences between the two conferences, but their overlap. And the most startling overlap area to me this week was discussion of Smart NIC.
On the surface, smartNICs don’t appear to be particularly important innovations. They’re literally improved networking devices in which you can do firewall segmentation and additional segment control points in the network interface before the traffic enters or leaves the host server. But they’ve been evolving to be more like a supervisory co-processor. That’s what we hear about AWS Nitro and it’s how VMware positioned Project Monterey’s relocation of the VMware hypervisor.
The simple reality is that AWS’ deep use of that technology as a supervisor is proving it’s value and driving the industry forward to keep up. That means we’re looking at a significant evolution in server architecture. At RackN, we are world leaders in multi-vendor server management and we know first hand that managing servers is already bespoke and tricky. Adding a dedicated supervisor system could really expose new controls; however, it also presents new management challenges.
I believe that this is the next generation of baseboard management controllers (aka BMC): they are basically a whole other computer that coordinates the functions in the main computer including access to virtual machines, storage devices, networking devices and graphics devices embedded in the system. This type of vendor attention indicates that the consensus that supervisors are required to manage infrastructure at scale.
It’s possible that we’re on the cusp of a new evolution in server architecture where new supervisory systems improve performance, manageability, security. If this can reach the mainstream in an integrated and manageable way, then we may actually have servers that significantly rebalances the choice between owning and outsourcing infrastructure management.